Update discussion of serialisation vs compression in CIP-0381

[Migrated from PLT-8797]

CIP-0381 contains a discussion of the costs of deserialisation and uncompression for curve points in the blst library. It looks as if this was based on erroneous benchmark results: deserialisation seemed to be an order of magnitude faster than uncompression, but it now seems that it’s more like 20%. This won’t affect our conclusions, since it means that deserialisation is even worse than we’d previously thought (it gives you a modest time improvement but incurs a large space penalty. We should try to be precise though.